The US Executive Order on AI
It isn't possible to summarize the EO in a page, and so here are just the highlights.
On October 30, the U.S. President issued a sweeping and critical Executive Order (EO) to jumpstart AI strategy and information gathering within the U.S. federal government, indicating priority concerns and setting a trajectory for potential U.S. AI regulation to come. It isn't possible to summarize the EO in a page, and so here are just the highlights.
AI Expert Recruitment
Initiatives are outlined to (at least attempt to) greatly bolster U.S. federal capabilities in AI by attracting experts and streamlining high-skill, AI-related immigration processes.
AI Model Security
To get a grip on increasingly powerful foundational AI models, the following reports will start rolling into the Department of Commerce within 90 days:
AI models created with compute power exceeding 10^26 flops must be reported, along with their safety protocols and results. This targets systems greater than the sophistication of OpenAI's state-of-the-art GPT-4, but well within the anticipated range of yet-to-be released models like GPT-5 or Google’s Gemini.
Data centers capable of 10^20 integer operations per second are to report their capabilities. No data centers of this size are publicly known, but most major tech firms are rumored to have data centers with this capacity.
Data centers offering Infrastructure as a Service (IaaS) at certain large scales must report when offering those services to international clients, following KYC-style standards.
Additional Broad AI Objectives
A broad spectrum of orders encompass actions related to AI's role in areas including critical infrastructure, cybersecurity, privacy, equity, labor, misuse prevention, fraud detection, content identification, and (perhaps most importantly) synthetic biology supply chain safeguards.
Regulatory Forecasts
Here are a few potential regulations for which the U.S. government may attempt to find technical toeholds, looking at the array of reports ordered in the EO:
Stringent safety protocols for AI model development and deployment
Further restrictions on foreign access to computational power, and restrictions on advanced AI services or technologies
Watermarks for AI-generated materials
Widespread adoption of privacy-enhancing tech
Anti-discrimination measures within AI functionality
Job protection mechanisms in the wake of AI integration